Bootstrapping a Bell Curve
The bell shape in bootstrapping comes from the sampling distribution of the mean, not from the underlying dataset — and that distribution becomes bell-shaped by the Central Limit Theorem as resamples grow in size.
A complete interactive classroom, not just a preview.
Start when you are ready to enter this Stage's 8 scenes and explore, respond, and learn as you go.
How does bootstrapping let you build a bell curve from a single messy dataset?
Your raw dataset looks nothing like a bell — it's lumpy, skewed, or too small. Yet the bell curve still appears. Watch how.
If one sample is messy, how can a smooth, symmetric distribution reliably emerge from resampling it over and over?
A side-by-side comparison of the original messy dataset against a distribution of sample means built by repeated resampling, plus a manipulable simulation that lets the learner change sample size and number of resamples.
Bootstrapping doesn't manufacture normality out of nowhere — it buys normality by averaging, and the bell belongs to the distribution of statistics, not to the original data.
Bootstrapping must somehow smooth out the original messiness directly, so the resulting distribution mirrors the shape of the source data.
- Confidence interval methods beyond the percentile approach
- Bayesian bootstrap
- Parametric bootstrap
- Bootstrap for regression coefficients in depth
- 01A Single Messy Dataset, Many Possible Bells?slideQuestion
Introduce the driving question by showing a skewed, lumpy dataset and asking how a smooth bell can be extracted from it by resampling alone.
- Original data is far from bell-shaped
- Resampling alone seems unable to invent symmetry
- Where does the bell come from?
- 02Commit Before We ResamplequizPrediction
Ask the learner to predict whether the bootstrap distribution of the mean will mirror the messy source distribution or look bell-shaped, before any evidence is shown.
- Make one explicit choice
- Commit to an intuition about shape vs. symmetry
- 03What Resampling Actually ProducesslideEvidence
Show the empirical fact: the original data stays messy across resamples, but the histogram of sample means becomes smooth and bell-shaped.
- Every resample preserves the source's lumpy shape
- The distribution of means is smooth and symmetric
- Two different objects are being plotted
- 04Resample and Watch the Bell EmergeinteractiveEvidence
Let the learner draw bootstrap resamples from a skewed dataset, compute the mean of each, and watch the histogram of means converge toward a bell as more resamples accumulate.
- Source dataset stays skewed on every draw
- Means accumulate into a bell shape
- Shape sharpens as the number of resamples grows
- 05Why Averaging Buys You a BellslideExplanation
Explain the Central Limit Theorem: each resample's mean is an average, and averages of independent draws tend toward a normal distribution regardless of the source shape.
- Each mean is an average of n draws
- Averages smooth out individual irregularities
- The bell belongs to the mean, not to the source
- 06When the Bell Fails to AppearslideBoundary
Show the limits: when the sample size is tiny, the source is heavy-tailed, or the statistic is a median or extreme quantile, the bootstrap distribution can stay lumpy and non-bell-shaped.
- Tiny resample size breaks the averaging effect
- Heavy tails slow convergence
- Medians and quantiles average less cleanly
- 07Apply the Idea to the MedianinteractiveTransfer
Test transfer: have the learner run a bootstrap on the median of the same skewed dataset and compare its distribution to the bell they saw for the mean.
- Median distribution is narrower than the mean's
- Median distribution is less bell-shaped and more stepped
- The bell was a property of the averaging statistic
- 08The Bell Belongs to the StatisticslideResolution
Resolve the driving question directly: the original dataset never becomes bell-shaped, but the distribution of its resampled means does — by the Central Limit Theorem applied to averaging.
- Source data keeps its messy shape
- Resampled means form a new, bell-shaped distribution
- Bootstrapping reveals uncertainty in a statistic, not in the data
Discussion threads for a Stage aren't available yet.
